Much has been written, discussed, and argued about AI replacing photography and photographers. I have participated in some of these discussions by writing or replying to comments. In general, the conversations primarily focus on how realistic the AI-generated “photographs” may look, except for the hands.
It is important to note that in this article I will focus on how images are created, not how they are edited, possibly with the help of AI like noise reduction, enlargements, and the like. We all use many tools to edit, clarify, sharpen, color-correct, and do many other things after image creation. Many forms of technology have helped photographers to achieve the desired results, and we are thankful for that. Computational photography tools are available either built into cameras, post-processing software or as separate applications.
My intention in this article is not to denigrate AI imagery but give it and photography their proper contexts. Artificial intelligence will likely continue to play a significant role in photography and provide alternative imaging possibilities when photographic qualities like connectedness to the real world are not necessary.
Table of Contents
Photographs vs AI
I would like to approach the subject from a different angle by taking a closer view of the characteristics of photographs and photography rather than comparing it to the potentially high quality of AI-generated images. The photographic process and the inherent and inseparable connection between photography and objects photographed need to be brought forward and emphasized.
Light vs Data
The word photography comes from the Greek, “photos” and “graphos.” It means light writing, or writing with light. In any form, a photograph records the reflected or emitted light from the real world on light-sensitive media. This is true in studio or landscape, astronomic or macro, and visible or extreme spectrum photography like infrared or ultraviolet. This is the nature and foundation of photographic image creation or photography. The recording medium may change from wet plates to sheet film, from roll film to digital sensors. But the light-sensitive nature of the medium remains.
AI-created imagery, on the other hand, is data-driven. Using special algorithms, AI technology studies the vast collection of images, and written words, and seems to “memorize” patterns in words or in images. Given a specific description of an imaginary idea, it dives into its database and memorized patterns to assemble images that reflect the words and sentences given to it. There is no light, or “photos” in the AI-generated images, lacking the fundamental element, the essence of photography.
Of Something vs Constructed
Photographs are always “of something” although they may also connect us to what it is about. A photograph always records a subject or subjects. The simple question may often be asked “what is it,” although an extension of it may also be relevant, “what is it about”?
This unbreakable link to the real world and strong ties to the things photographed are not present in AI imagery. The images created by AI technology are “of nothing” beyond the trigger words the image creator may utter. It is akin to a massive fishing expedition to make the AI find relevant images in its database and construct a realistic image based on those. They lack the connectedness of photography to the real world and real subjects.
Analysis vs Synthesis
Photography is an analytical art form. Photographers look around, analyze the scene, and select one frame or a few from among an infinite number of frames. Ideally, it reflects our seeing. We can certainly raise the camera, lower the camera, rotate it, pitch it, yaw it, aim at a different part, and end up with a different photograph.
This is distinctly different from painting or drawing which begins with a blank canvas and synthesizes the final art form by adding shapes, colors, forms, etc. with utmost freedom to place things where the painter wants. Perhaps that is one of the strengths of painting. Take a look at the pencil drawings of Diego Fazio like the one on the right and you may think they are photographs. But they are hyperrealistic drawings by a talented artist.
Because of the shared additive synthesis, AI images may be closer to painting than photography. They may be closest to hyperrealistic paintings that resemble photographs, images created by NI (natural intelligence), and are not called photographs.
Seen vs Imagined
In some ways, seeing is the essence of photography. While performing this analytical art form, we aim our lenses to specific parts of the world around us to pick a frame because, in our analysis, that particular frame presents the photograph we wish to take. Ideally, it reflects our seeing.
Seeing is a mental process and different from looking. It starts with our eyes capturing some information from our surroundings but continues as a mental activity. It requires awareness, and awareness is a state of mind. There are many things you are looking at now, but not necessarily aware of them. We see things in our ways that make us take different photographs in the same place.
AI, on the other hand, produces imagined scenes with no seeing or frame selection involved. The imagined constructs are narrated to the artificial intelligence device which harvests its large database to create a composite. It is the surprising quality of the synthesized images that mesmerizes the viewers and makes them appear as a substitute for photography. Keep in mind that AI technology is not limited to realistic photographic image creation, and I am sure it can easily create images of paintings or drawings of various styles given a similar set of instructions.
Encoded vs Decoded
At least some photographers try to encode messages by way of photographic structure and language. There are many elements of the structure, and when properly brought together, they help convey a message. By that, I don’t mean something profound, but a meaning that can be decoded by the viewer for a shared experience. That is the essence of communication.
AI has a strong hand in decoding visuals in its database. However, the resulting image is not structured by the image narrator by selecting some available elements or angles of something they see. The AI user creates an image in his or her mind, not much different from that of a novelist, and explains this imagined view to the machine. The AI device first decodes the words, then the material in its database, and synthesizes an image that visualizes the narrated scene. An iterative process may follow between the image visualizer and the AI device.
Journalistic vs Hyperbolic
Photography can and does capture things as they occur. Recording of current events on the streets or on stage cannot be done truthfully by AI systems. Photography remains the primary tool for capturing events as they happen from the angles photographers choose. Photojournalists are still the primary source of visual capture and information before any image creator and AI can even begin to think about the events unless, of course, we are interested in fakery.
Perhaps after a period of time and accumulation of visual and verbal stories, artificial intelligence may attempt to recreate some events if some potential inaccuracy is acceptable. AI needs a passage of time before it can start utilizing past information.
Actual vs Limited Commercial Use
There have been great hopes (or fears) that AI imaging will replace all the photographers in the advertising and commercial photography fields. I have some doubts about that although many may disagree with me. Please keep in mind that I am not talking about using AI to reduce digital noise or many other similar uses. I am sure there are areas in commercial photography AI may already be in use but, it may eventually hit a brick wall as some legal requirements are imposed upon the images.
For instance, a concept called truth in advertising prevents using substitutes in some product photography. Yes, if you are photographing for a diner menu, you may be able to use Elmer’s Glue instead of milk in the bowls. But the same substitution cannot be made if the image is for a brand of milk. That needs “real milk” to be used and photographed. Likewise, using substitutes for ice cream instead of the real thing may be acceptable for a restaurant billboard, but not for promoting a brand of ice cream.
If these norms are to be maintained, and I have only scratched the surface, how can AI imagery be used in advertising with nothing in them being real? Time will tell, but I would not bet against the regulatory institutions that protect the consumers’ interests.
A Simple Experiment
I have no experience in AI image creation but did a quick search on creating AI images online. Not knowing which one to choose, I picked a couple, more or less at random. Then, I even tried the Open AI, Dall-E. But, their system was continually busy and I could not get any results from them.
The following are the results produced for two different narratives. I can photograph one, the second one is highly unlikely.
Description One
“Turkish prayer beads hanging on the edge of a copper bowl.” And a variant “Turkish prayer beads in a tin-plated copper bowl hanging on the edges.” These resulted in the following on Fotor and Picsart.
Click on the images to see them larger, uncropped, and read their titles.


Description Two
“One-legged old man riding a bicycle uphill.” The alternate “One-legged old man riding a bicycle climbing the hill.” Yielded the following sets.


The second attempt shows the tenacity of artificial intelligence engines. One even created an image with a bicyclist flying in the air and another with what appears to be an amputated leg sticking out. The other engine created awkward old men on bikes with hints that they may only have one leg, and other missing pieces from the images. Yes, there are bicyclists with one leg and they can climb hills, but the databases used by these artificial intelligence image creators may not contain their information yet. And, they are young and athletic, not old men. After my creations, these engines may have learned some things about the concepts I described.
Finally With Dall-E After Many Attempts
As I said earlier, my attempts to access Dall-E had been futile as it was constantly busy. But today, after about 5-6 days of attempts, I managed to access it and generated the following images. For some reason, “one-legged” did not seem to register and Dall-E depicted the old men primarily from behind. Oh, climbing a hill also had varied meanings. The captions include my instructions.






Before you find faults with my method of creating AI images, allow me to say first, that I wanted something searchable yet with some complexity and nuance. Second, I wanted to create an image that would be impossible for me to photograph. As you can see, in the first instance there is nothing that resembles “Turkish prayer beads” but offers an anime woman in some kind of costume and some pots.
Yes, my attempts may be far too simple and unfairly oxymoronic. I wanted to see the level of knowledge of these technology tools and their willingness to create an exceedingly unlikely scenario. In more experienced hands the results could have been different. But, don’t be taken by the apparent high quality of the images, which I found seriously lacking in my results.
The question still remains how well can AI depict designer fabrics made into new lines of clothing worn by believable models with accurately rendered hands?
Some Technical Questions
I am truly curious and asked these questions elsewhere. There were no replies to any of the questions. It may be important that we understand the answers to the following before continuing the emergent love affair with AI imagery. I asked these questions to the siblings, or cousins of AI-image generators, ChatGPT and Google Bard. They are in the supplementary articles published in tandem. You will find the links below.
- What format is the AI-created digital file? JPEG, TIFF, DNG, PSD, …? Can it be any other format?
- Is there an associated color space and bit depth embedded? Or, is it a RAW file?
- How is the file created, pixel by pixel, section by section, or all at once?
- Do the pixels have Red, Green, and Blue sub-pixel elements with Bayer filter patterns?
- If there is a Bayer pattern, what purpose does it serve?
- How big is the file, or is it dependent on the instructions? How big can it be?
- Is there a particular aspect ratio or does it depend on instructions?
- Does AI create or can it create EXIF records?
- Is there any “camera” and exposure information saved?
- Can the image be identified as AI-generated by looking at the above and other file specifics?
For someone who has never tried creating or has seen a first-generation AI image, these are nagging questions. I will welcome reasonable explanations, they will be highly educational for me.
I believe AI will not ruin photography and photographers by becoming a substitute for them. It may ease the lives of photographers and more easily correct the perspective, but a photographer will still take pictures of architectural marvels before AI can attempt to collate them later for other uses.
AI Perspectives On These Issues
Before I published this article, I had a few chat sessions with AI engines starting with ChatGPT and continuing on Google Bard. The conversations were eerie, entertaining, informative, and revealing some features of AI chat engines or lack thereof. The text of those sessions is quite long. In order to keep this article at a somewhat reasonable length, I am presenting them in parallel articles that interested visitors may follow.
Related Articles
I have used in this article parts from my previous writing. For further reading, I may recommend the following among others that explore the essence of photography:
Andrew Webb
I think you are going off half-cocked. Learn how the different AI models work, and learn how to use them before speculating and pontificating.
Haluk Atamal
Cemal, it took me quite a long time to respond to your post, I know. That was both because I was unnecessarily busy at this time of the year and the article needed some serious pondering upon.
First I like your meticulously and very carefully chosen wording for a subject open to a lot of debate. From my point of view, I agree with all of them. I especially liked the comparisons for light, journalism, and the technical questions.
Special note for Mr. Webb’s very open hearted comment: I have no idea how different AI models work and have no experience on how to use them. On the other hand, I still feel that he is disclosing some sort of apprentice-like vision by his very collection of the words used. That, by the way, is only a feeling and should have no consequences to his generous openness.
To be frank, I would love to read a similar post from him, from his viewpoint.
Take care and stay well. Rgds to Jan!
Haluk
A. Cemal Ekin
I was beginning to worry, Haluk! I am glad you found the time to write a comment. Mr. Webb may ponder his comment sometime in the future. I think this love affair with any form of AI may soon get regulated as it should. I also think intelligence cannot be separated from senses, feelings, and emotions. How do you learn (become more intelligent) that touching a stove or a hot iron burns? What does it feel like to cut your finger? And many more feelings and emotional matters are a part of natural intelligence.
But, on the matter of photography, unlike some meaning stretchers, AI-generated images will look like photography but not be photography.
Good to hear from you,
Cemal
Andy Barnham
Thanks for this Cemal. Could I also, please, offer the following, in relation to your Seeing vs Imagined section? While I appreciate the notion of happy accidents, I am very deliberate in what I photograph and how in regards to the taking of the photo, ie the composition (what is included and what is not), the exposure, the focus and also post production, ie the editing, colour correction, exposure, curves, contrast etc… While I have been told I have a signature style (ref my editing), a lot of my editing originates from how I feel I should present an image. All of what I have just talked about can be surmised in one word; control. I have little to no control over the AI; even if it has learned off a data set which I have submitted, there is a lack of finesse and control over the result.
A. Cemal Ekin
Andy, thank you for stopping by and sharing your thoughts. Our approach to photography seems to have many common threads. I edit my images to reflect what I saw and what I felt. You may find some articles I wrote here that convey the same idea. And, I agree that AI as a tool does not give control to the creator like doing it yourself. I think the primary reason may lie in AI not having any feelings or emotions, it does not know getting burned after touching a hot iron or cutting your finger. Without emotions, images will be lacking.
Take care, and stop by again,
Cemal